From mboxrd@z Thu Jan 1 00:00:00 1970 Received: from [140.186.70.92] (port=57636 helo=eggs.gnu.org) by lists.gnu.org with esmtp (Exim 4.43) id 1OGoYh-0005OK-LK for qemu-devel@nongnu.org; Tue, 25 May 2010 03:28:16 -0400 Received: from Debian-exim by eggs.gnu.org with spam-scanned (Exim 4.69) (envelope-from ) id 1OGoYg-0000yb-FB for qemu-devel@nongnu.org; Tue, 25 May 2010 03:28:15 -0400 Received: from mx1.redhat.com ([209.132.183.28]:33775) by eggs.gnu.org with esmtp (Exim 4.69) (envelope-from ) id 1OGoYg-0000yT-7C for qemu-devel@nongnu.org; Tue, 25 May 2010 03:28:14 -0400 Message-ID: <4BFB7BFE.9030707@redhat.com> Date: Tue, 25 May 2010 10:27:58 +0300 From: Avi Kivity MIME-Version: 1.0 Subject: Re: [Qemu-devel] [PATCH] resent: x86/cpuid: propagate further CPUID leafs when -cpu host References: <1274428240-10801-1-git-send-email-andre.przywara@amd.com> <4BFAF941.1010604@codemonkey.ws> In-Reply-To: <4BFAF941.1010604@codemonkey.ws> Content-Type: text/plain; charset=ISO-8859-1; format=flowed Content-Transfer-Encoding: 7bit List-Id: qemu-devel.nongnu.org List-Unsubscribe: , List-Archive: List-Post: List-Help: List-Subscribe: , To: Anthony Liguori Cc: Andre Przywara , qemu-devel@nongnu.org, aurelien@aurel32.net On 05/25/2010 01:10 AM, Anthony Liguori wrote: > On 05/21/2010 02:50 AM, Andre Przywara wrote: >> -cpu host currently only propagates the CPU's family/model/stepping, >> the brand name and the feature bits. >> Add a whitelist of safe CPUID leafs to let the guest see the actual >> CPU's cache details and other things. >> >> Signed-off-by: Andre Przywara > > The problem I can see is that this greatly increases the chances of > problems with live migration since we don't migrate the cpuid state. -cpu host is already problematic for live migration. Are you talking about the state maintained by the cpuid instruction? Yes, we need to migrate those bits. > > What's the benefit of exposing this information to the guest? > Some algorithms adjust themselves based on the cache size. If you have several passes over a large data set, it's often better to run each set of passes on a subset of the dataset that fits in cache, then stitch the subsets together. -- error compiling committee.c: too many arguments to function